1

Chinese Competition and its Effects on Mexican Maquiladoras

André Varella Mollick* Karina Wvalle Vázquez**

* Mollick: Corresponding Author. Department of Economics and Finance, College of Business Administration, University of Texas - Pan American, 1201 W. University Dr., Edinburg, TX 78539-2999, USA. E-mail: [email protected] Tel.: +1-956-316-7913 and fax: +1-956-384-5020. ** Wvalle: EGADE, ITESM-Campus Monterrey, Av. Fundidores y Rufino Tamayo, San Pedro Garza Garcia, N.L., 66269, Mexico. E-mail: [email protected] Tel.: +52-81-8625-6214 and fax: +52-81-8625-6028. We are grateful to helpful comments and suggestions from Joan Anderson, Jorge González, Jorge Ibarra, Nicolas Sisto, and two anonymous referees of this journal, who are not, however, responsible for any remaining errors in this article. Initial stages of this research were accomplished under the “Research Seminar on the Border Economy” held between January and May of 2004 sponsored by the research chair “The Economic Agenda of the Mexican Border” at the ITESM-Campus Monterrey in Mexico.

2

Chinese Competition and its Effects on Mexican Maquiladoras

Abstract: Contrary to popular press reports, this paper provides evidence against the idea that Mexican maquiladora employment falls substantially with higher (relative) MexicoChina wages. Applying panel data and seemingly unrelated regression (SUR) system methods on Mexican maquiladora employment across border states between 1990 and 2001, negative wage effects are found in levels. The stationary residuals of these estimations are employed in a two-step (panel) cointegration procedure. Under the errorcorrection form, these residuals are statistically significant and the effects of relative wages on maquiladora employment vary between -0.076 and -0.091. The effects of U.S. real output growth, however, are very strong, with coefficients varying from 3.548 to 3.652. While wage coefficients do not change much from earlier research, the magnitudes of U.S. output coefficients support anecdotal evidence based on: “When the U.S. sneezes, Mexico catches a cold”. Compared to previous research, our results reflect the increasing integration between the two economies.

Keywords: Border, China, Employment, Maquiladoras, Mexico. JEL Classification Numbers: F14, F16, R12.

3

1. Introduction The average real value added growth of Mexican maquiladoras (export assembly plants) has been around 10% in the period between 1990 and 2002, much higher than the 3% real growth of Mexican GDP in the same period, according to Hanson (2002). Maquiladoras are now responsible for about half of Mexican exports (up from 15% in 1980), according to Kose et al. (2004). Hummels et al. (2001) indeed conclude that vertical specialization (the value of a country’s imports embodied in its exports) has played an important role in the growth of Mexican exports since 1979. After all, the degree of integration between inputs and outputs of this industry leads to its operating mechanism: “Maquiladora firms are those that import nearly 100% of their inputs and then export nearly 100% of their output” (Robertson, 2003, p. 40). What explains employment growth at Mexican maquiladoras? Some analysts have observed that recent “structural changes” in the industry requires more than the U.S economic situation and sharper competition, mainly from China, as driving forces. Banamex (2003), for example, notes that there are states in which maquiladora production are booming, such as Nuevo Léon or Jalisco, despite the recent recession. And there is higher male intensity in the industry, which suggests less importance of labor costs as male workers earn more than female ones, together with a growing importance of “other manufacturing”: precision instruments for medical use, jewelry and craftwork. Despite these structural changes, there is a general feeling that China is growing too fast, selling too much abroad and, possibly, becoming a threat to economies with which it competes. Bernard et al. (2005) examine the relationship between import competition from low wage countries and the growth of U.S. manufacturing plants. Less formal studies, such as The Economist (2003), refers to “what the rest of the world fears most about China – that its phenomenally fast growth can be sustained only at the expense of other economies,

4

both developed and developing”. On estimates of relative wages between Mexico and China, in particular, The Economist (2003) reckons that China is “by far the biggest garment exporter in the world, with average wages in the industry of 40 cents an hour - less than a third of, say, Mexico’s… With manufacturing wages in China averaging about 60 cents an hour - 5% of the American average, and 10% of that in some neighboring Asian economies.” Others regard the relatively strong peso of the early 2000s as the major factor: “Mexico’s labor costs climbed about 50% in dollar terms as the peso held at about 9.5 per dollar between 1999 and 2002 and inflation accelerated faster than in the U.S. Mexican maquiladora workers are now paid about $50 to $75 a week, a third more than their counterparts earn in China.” (Milwaukee Journal Sentinel, 2003). Due to Chinese competition, signs of political unrest have been noticed in Mexico. There have been reports of “textile and shoe workers … thrashing Chinese goods in the streets. The government has started airing ‘Buy Mexican’ ad campaigns, and police have rounded up Asian vendors and staged increasingly violent raids against street stalls selling contraband imports… In one recent raid in Mexico City, police rounded up Koreans – who allegedly run many of the import operations – and deported 11 of them, drawing complaints of discrimination from the Korean community.” (Nation & World, 2003) Some sectors are particularly affected. Shoemakers, for example, “complain they are being driven out of business by cheap Chinese imports. “We just can’t compete with the labor costs”, said a project director for Mexico’s Apparel Industry Chamber. “Labor in China costs 48 cents per hour, and in Mexico it’s $1.20.” (Nation & World, 2003) While currency appreciation should diminish the aggressiveness of China’s expansion abroad, the revaluation of the yuan (fixed at 8.3 per USD since 1995) is still an entirely open issue. On the causes of the Mexican maquiladora expansion, Hanson (2002) lists four factors: First, the ability of multinational firms to fragment production across borders

5

through outsourcing; second, Mexico’s low wages relative to the rest of North America; third, trade policies that gave maquiladoras special advantages in exporting to the U.S. market; and fourth, Mexico’s geographical proximity to the large and rich U.S. economy. Cypher (2001, p.27) contrasts the rapid growth of Mexican maquiladoras to the declining manufacturing sector as one of the pillars of his “disarticulation thesis”: “A vicious circle is established whereby the comparative advantage of cheap labor (without independent union representation) becomes the mainstay of a stop-and-go Mexican economy.” We propose an econometric model in this paper that captures some of these elements. In doing so, we build on previous econometric studies of employment and output dynamics of Mexican maquiladoras. The papers by Fullerton and Schauer (2001) for the El Paso-Juárez border city and Mollick (2003) under panel data across Mexican states contain estimates of employment equations. Gruben (2001) presents a time series model for annual data from 1975 to 1999 with U.S. industrial production affecting positively maquiladora’s aggregate employment (coefficients ranging from 1.05 to 1.31) but does not find the NAFTA dummy variable as statistically significant. He also includes the ratio of hourly Mexican manufacturing wages with respect to a group of 4 Asian countries and obtains a statistically significant coefficient (at the 10% level) of about -0.11: as Mexican wages rise with respect to Asian, labor demand in the Mexican maquiladora falls slightly. Similar to these studies, this paper employs a simple model with relative wages and U.S. real GDP growth as major forces behind maquiladora employment growth. A NAFTA dummy variable also measures any structural change due to the signature of NAFTA and its implementation from January 1994 onwards. However, the major novelty here is the view that a potential foreign investor in Mexico or China calculates labor costs and internalizes this piece of data in his or her information set when deciding in which state of the Northern Mexican border to invest.1 This idea is grounded on the multinational

6

enterprises (MNEs) literature, such as: Dunning (1988) and Markusen (1995). More recently, Mortimore et al. (2003) refer to Mexico and the Caribbean basin as examples of the bulk of FDI coming from efficiency-seeking MNEs which set up export platforms in this subregion as part of their regionally or internationally integrated production systems. We hereby calculate a relative wage variable in manufacturing between Mexico and China and exploit its role in feasible generalized least squares (FGLS) estimates under panel data, as well as in system-based seemingly unrelated regression (SUR) estimations. Contrary to speculation in the press, this paper argues that there are no sound empirical arguments to justify that maquiladora employment levels fall substantially with higher (relative) Mexican wages. This is so because the only strong negative effects found in this paper are confined to preliminary estimations in levels. We next employ the (stationary) residuals from one of the level estimations in a two-step (panel) cointegration based procedure. Under the first-differenced, error-correction form, the effects of relative wages vary between -0.076 and -0.091, which are close to those reported by Gruben (2001). Our results are thus consistent with stories of U.S. investors operating in a global world, such as Huro (2003) and Fullerton and Barraza de Anda (2003), in which labor costs are weighted against other forces, such as: distance, infrastructure, and set up business costs. This research confirms, on the other hand, the very strong role of U.S. business cycles in the labor growth of the Mexican maquiladora industry. The effects of U.S. output growth are always strong, with coefficients varying from 3.548 to 3.652. These values are higher than the results in Gruben (2001) for the aggregate of Mexican maquiladoras, who omitted the years associated with the U.S. latest recession. This paper comprises five sections. Section 2 explains the data construction and sources and section 3 contains the applied empirical methodology. Section 4 presents the major results of our work and section 5 contains extensions for further research.

7

2. Data Description In order to explain employment (L) dynamics of Mexican maquiladoras, this study examines real relative wages between Mexico and China (WMXCH) and U.S. real GDP (Y*) as major economic variables. We will also employ real relative wages between Mexico and the U.S. (WMXUS) as an alternative explanatory variable. When observing the relative change that has occurred in manufacturing wages between Mexico and China from 1990 to 2001 we look at total average remunerations per month for all type of workers.2 Chinese data wages are collected from the International Labor Organization (ILO: http://www.ilo.org) and Mexican wage data are obtained from the Monthly Industrial Survey provided by the Instituto Nacional de Estadística, Geografía e Informática (INEGI: (http://www.inegi.gob.mx). Consumer price indices for both countries are taken from the International Financial Statistics (IFS: http://ifs.apdi.net/imf/logon.aspx) of the International Monetary Fund (IMF). Whenever needed, the spot exchange rate comes from Banxico (http://www.banxico.org.mx) and refers to average quotes in a month. These two datasets on Mexican and Chinese wages contain the same definitions of wages. They include the average monthly wage per year for administrative workers and for production workers in the manufacturing industry. The time dimension, however, is not the same. For China, the data available are total average wages per month at annual frequency, whereas for Mexico the data are in monthly frequency of total earnings. Therefore, we aggregate the Mexican data in order to compare them to Chinese data. In Mexico, the original data are in total monthly earnings of the personal occupied in the manufacturing industry, which is averaged out for each year. The data are in U.S. dollars (USD) so as to compare monthly average total earnings per year between countries. Figure 1 shows Mexico-China real relative wages (WMXCH), plotted against the multilateral Mexican peso (q), calculated by Banxico with respect to 111 currencies around

8

the world and deflated by consumer price indexes. There is a real appreciation of the peso prior to the December 1994 peso crash, a pattern well documented in the currency crisis literature. The December devaluation makes the real exchange rate respond slightly in 1994, followed by the enormous impact in 1995. Since then, the peso have been appreciating in real terms against the multilateral basket of currencies, ending 2001 at its minimum of 62.93, a year marked by the BANAMEX purchase by U.S. based-Citicorp and record-high FDI inflows. Figure 1 makes clear that average relative Mexican wages (WMXCH) in 1990 is 16.45 times (right axis) Chinese wages. This behavior of WMXCH remains essentially unchanged until 1993. In 1994 there is an increase due to the depreciation of the yuan exchange rate against the USD: from 5.76 in 1993 to 8.62 in 1994, while the peso changed abruptly in December of 1994, as the average peso exchange rate moved from 3.12 to 3.38. In 1995, however, the yuan remains fixed while the peso feels much of the adjustment of becoming a floating rate currency. In 1995 the Mexican peso falls to 6.42 pesos per USD (annual average), as can be confirmed by inspection of table 1. In the late 1990s, there is a downward trend in the wage ratio between Mexico and China. This trend can be explained by the rigidity of the yuan and the smaller rate of adjustments of the (now floating) Mexican peso: Mexican peso yearly averages are 9.56 in 1999, 9.46 in 2000, and 9.34. The final relative wage ratio for 2001 is 3.94, which makes Mexican wages about four times Chinese wages paid in manufacturing. This figure appears to be roughly of the same magnitude of informal discussions on Mexican-Chinese relative wages, such as those appearing in the popular press and refereed to in the introduction: The Economist (2003), Milwaukee Journal Sentinel (2003), and Nation & World (2003). Although the main wage variable in this paper is WMXCH, it is also useful to contrast Mexican with U.S. manufacturing data. On the Mexican/U.S. wage data, WMXUS is taken directly from the U.S. BLS publication “Hourly Compensation Costs for

9

Production Workers in Manufacturing”, except for the WMXUS figures for 1991 and 1992 that were linearly interpolated, based on the actual figures for 1990 and 1993. In the BLS data, the compensation measures are computed in national currency units and are converted into U.S. dollars at prevailing commercial market currency rates. Hourly compensation costs in the BLS data include: i) hourly direct pay; and ii) employer social insurance expenditures and other labor taxes. Figure 2 shows the increase in Mexican wages before the Peso crash, the fall in Mexican wages in 1995 with the peso depreciation, the flat pattern in the mid-1990s, and some recovery in the early 2000s. The multilateral real exchange rate is the same as in figure 1. Mexico’s INEGI contains representative surveys of Mexican manufacturing. INEGI’s Banco de Información Económica (BIE) dataset refers to Employment (L) as all people employed during a month in a given establishment or outside it (under the establishment control), under regular compensation schemes. It includes regular workers under temporary leave and excludes workers that provide temporary work or retired workers.3 The model below includes fluctuations in the U.S. economy as major forces behind maquiladora employment dynamics. U.S. real GDP in USD (Y*) and the U.S. level of interest rates (i*) are obtained from Mexico’s INEGI in the “Indicadores Selecionados” section. We take monthly observations and then calculate the arithmetic average for each year in the sample. Figure 3 displays the flat U.S. output during the recessions of 1991 and 2001 and the rising trend from 1992 onwards, together with the U.S. T-bill interest rates at bottom 3% levels during 1993 and again with the more recent U.S. economic downturn. The six states of the Mexican Northern border with the U.S. are: Baja California, Chihuahua, Coahuila, Nuevo León, Sonora and Tamaulipas. This definition is based on official publications since the physical border that exists between the state of Nuevo León

10

and the U.S. is very narrow. Figure 4 contains the rising trends in employment, with some reduction in 2001, coinciding with the U.S. recession. For the last year of the sample, maquiladora employment at the border state of Chihuahua reaches 289,185, or 26.06% of the total. Measured as averages during the 1990-2001 period, maquiladora production of the 6 border states reaches 90.7% of the total output, of which Chihuahua is responsible for 28.6% of total maquiladora employment, Baja California for 25.3%, and Tamaulipas for 18.6%. Detailed shares of employment and production are available upon request.

3. The Framework The decision to invest in developing nations is based on several economic and institutional factors. The literature on MNEs, in particular, contends that such firms’ operations arise due to the fact that these firms possess some special advantage such as superior technology or lower cost brought about by scale economies. See Markusen (1995). Dunning (1988) has proposed the ownership (O), location (L), and internalization (I) views, which came to be known as the OLI-framework. Ownership advantage could be a production process to which other firms do not have access, such as a patent. The foreign market must offer a location advantage that makes it profitable to produce in the foreign country rather than simply produce it at home or export it. Examples are tariffs, quotas, transportation costs and cheap factor prices. Finally, the internalization advantage handles the production process abroad as being exploited internally by the firm rather than at arm’s length markets. The model below contains some of these elements. Let the general framework be as in Milner and Wright (1998) and recently applied to the Mexican maquiladora industry by Mollick (2003). Assume that maquiladora firms operating in each Mexican Northern border state (indexed by i) share a Cobb-Douglas production function as follows4:

11

Yi = Aγ Ki α Li β

(1)

where: Y represents real output, K is the stock of capital, L is the number of employees, and A is a productivity factor, whose parameter (γ) allows for changes in the efficiency of the production process. The α and β parameters are factor share coefficients of the inputs. The profit maximizing firm in a given state (“i” is omitted henceforth) will employ K and L at levels such that:

MRPL = w = ∂Y/∂L = Aγ K α βL β−1

(2a)

MRPK = r = ∂Y/∂K = Aγ α K α−1 L β

(2b)

From (2b), we know that: Lβ = K r / (Aγ α K α), which substituted into (2a) yields:

K = (wαL)/(βr) = (w/r) (αL)/β

(3)

Substituting (3) into (1) yields:

Yi = Aγ [(αLi/β) (w/r)]α Liβ

(4)

Applying logarithms to both sides of (4) leads to the following equations:

Log Yi = γ log A + α Log α + α Log Li - α Log β + α Log (w/r) + β Log Li

(5)

12

Log Yi = (γ log A + α Log α - α Log β) + (α + β) Log Li + α Log (w/r)

(6)

Log Li = [-1/(α + β)] (γ log A + α Log α - α Log β) − [α/(α + β)] Log (w/r) + [1/(α + β)] Log Yi

Log Li = β0 + β1 Log (w/r) + β2 Log Yi

(7a)

(7b),

where: β0 = [-1/(α + β)] (γ log A + α Log α - α Log β); β1 = − [α/(α + β)]; and β2 = [1/(α + β)]. Assuming perfect capital markets, the real cost of capital fluctuates over time but not within states. We discuss below the treatment of time effects. Assuming the maquiladora firm has the option of investing either in Mexico or in China, or either in Mexico or in the U.S., its employment across Mexican states at the U.S. border would be a function of real relative wages (W). Given the information provided in section 2, W can be either WMXCH or WMXUS. Real U.S. GDP (Y*) seems to capture well the demand side without incurring the problem of simultaneity that is embedded in the Yi term of (7). Previous works with Y* as an exogenous variable in employment growth equations include Fullerton and Schauer (2001) and Mollick (2003). Any additional variables that affect the employment relation can also be incorporated at this stage.5 Introducing the cross-sectional maquiladora term (the state maquiladora effect: λi) together with the time series element (the time specific effect: δt), and the white-noise error term (εit), the equation to be estimated becomes6:

LogLit = λi + δt + β1LogWt + β2LogY*t + εit

(8)

13

Differencing (8) yield dynamic labor demand functions:

∆(LogLit) = α + β0i + β1∆(LogWt) + β2∆(LogY*t) + β3NAFTA + νit

(9),

where: α represents the time differential: ∆δt, β0i is allowed to vary across states (the fixed effects), and νit = εit - εit-1. The NAFTA dummy variable, defined as 0 for the years 1990 to 1993 and as 1 for the years 1994 to 2001, captures employment expansion due to the signature of the trade agreement as in Gruben (2001). A positive and statistically significant value for the NAFTA coefficient would suggest that NAFTA drives maquiladora growth, while a negative value would mean that NAFTA discourages employment growth in the maquiladora sector. We estimate (8) and (9) by the feasible generalized least squares (FGLS) fixed effects model, with seemingly unrelated regression (SUR) weights, which is the feasible estimator

when

the

residuals

are

both

cross-section

heteroskedasticity

and

contemporaneously correlated. We also use cross-section weights for robustness purposes. Tests of serial correlation in the residuals of (9) detect only a few rejections of the null hypothesis, which means the differencing procedure removes residual correlation in general.7 In addition to (8) and (9), we put forward an error correction model (ECM) that combines the short-run properties of statistical relationships in first differenced form with the long-run properties of the model in level form:

∆(LogLit) = α + β0i + β1∆(LogWt) + β2∆(LogY*t) + β3ECMit-1 + νit

(10),

14

where ECMit-1 are the lagged one period residuals from each (state) level equation (8): with only wages; with wages and real U.S. output; and with wages, real U.S. output and the NAFTA dummy. Following the cointegration literature initiated by Engle and Granger (1987), we expect the β3 coefficient in (10) to be negative: as deviations from the long-run equilibrium increase, the impact on the dependent variable decreases in an error-correction fashion. Equation (10) will be estimated in two forms. First, by FGLS methods as done in models (8) and (9); second by the SUR method which is appropriate when all the (RHS) regressors are exogenous and the errors are heteroskedastic and contemporaneous correlated. While in general equilibrium wages and employment are simultaneously determined, a large labor demand literature assumes wages are given when estimating employment functions. See Slaughter (2001) on the partial equilibrium approach to labor demand within international trade. U.S. real output, however, is clearly exogenous to maquiladora dynamics as well as any institutional variable such as the NAFTA dummy. In both methods (FGLS and system SUR), the lagged residuals from each equation (8) in levels are introduced into the (RHS) in an error-correction fashion.

4. Results Before the estimations, we explore the correlation coefficients for the series in levels and in first differences. In first-differences, for example, variations in relative wages (WMXCH and WMXUS) are positively correlated (0.823), while changes in relative wages are unrelated to U.S. output changes: -0.039 for WMXCH and 0.031 for WMXUS. Changes in Mexico-China relative wages do correlate strongly with changes in the Mexican peso spot rate (-0.857: the higher s, the lower is the ratio of WMXCH) and with changes in the real exchange rate (-0.736). The same holds for Mexico-U.S. manufacturing

15

wages: there is a strong correlation between WMXUS and changes in the Mexican peso spot rate (-0.921) and with the real exchange rate (-0.843). These are expected as the nominal exchange rate appears in the calculation of relative wages. Given these correlation patterns, we thus find it reasonable to exclude exchange rates from the set of independent variables as in (8). Table 2 below reports the fixed effect estimates of running several versions of equation (8) on data of annual frequency from 1990 to 2001 per state. We omit the results on equation (9) for space constraints and table 3 contains the ECM estimations associated with (10). The method of estimation for the panels in table 2 is the fixed effects model, with seemingly unrelated regression (SUR) cross-section weights. This is the feasible estimator when the residuals are both cross-section heteroskedastic and contemporaneously correlated. The weights and coefficients are continuously updated until convergence. Table 2 contains estimates of variations of equation (8) with the following independent variables: 8a) real relative wages only, 8b) real relative wages and U.S. output (Y*), and 8c) real relative wages, U.S. output, and the NAFTA dummy. The constant and fixed effects terms that vary across states are included in the estimations. As typically found in estimation in levels, the adjusted R2’s are close to one and misspecifications are found across the columns of the table. Specification (8a), for example, turns out to yield strongly negative coefficients, particularly for Mexican/Chinese wages: -0.496 in the equation with WMXCH and –0.358 with WMXUS. If correct, these findings would suggest the following story: as Mexican wages rise with respect to Chinese, employment at Mexican maquiladoras at states along the U.S. border falls. This turns out to be precisely the logic behind the popular press comments in the introduction on the substantial Mexican maquiladora employment effects caused by the relative wage impact.

16

Introducing real U.S. output in (8b), the real wage variable between Mexico and China becomes not statistically significant with a negative coefficient of -0.039. It is statistically significant, however, for the real wage variable between Mexico and the U.S.: -0.124. As relative U.S. Mexican wages (WMXUS) go up, employment at Mexican maquiladoras goes down as in the first specification. This would mean that the demand for labor at the maquiladoras is sensitive to changes in the relative wage rate between Mexico and the U.S. In (8c) this latter finding holds when the NAFTA dummy is included as the β1 is again not statistically different from zero (coefficient of -0.027) for WMXCH and is statistically significant (-0.072) for WMXUS at the 5% level. The effect of U.S. real production on employment is positive and strong as expected. Across models (8a) and (8b) the estimates vary only slightly: from 2.875 to 2.979 for the equation with WMXCH and from 3.016 to 3.108 with WMXUS. Whenever introduced, the NAFTA dummy variable is statistically significant, suggesting the institutional framework of North America per se help explain variations in Mexican maquiladora employment. The effect is positive under (8c) in table 2: 0.139 under WMXCH and 0.110 under WMXUS. This suggests that the timing after the signature of NAFTA treaty has some influence on maquiladora employment. This finding can be contrasted with previous works. Lack of results on the NAFTA dummy have been offered by Gruben (2001), while Mollick (2003) found evidence of negative coefficients for the panel of non-border states: the NAFTA dummy diminished employment growth in states away from the U.S. border. The latter is consistent with NAFTA favoring expansion of maquiladoras located along the U.S.-Mexican border, which is one of the results of this paper. Despite generally appealing from a theoretical perspective, these results are not without qualifications. Inspection of Durbin-Watson statistics and the more elaborate

17

formal tests on serial correlation indicate that all specifications in table 2 suffer from serial correlation problems either by the LM t-statistics or by the LM NR2 statistics.7 The null of no serial correlation is rejected for (8a), as well as in all other columns for models (8b) and (8c). In order to solve the specification problem, one has to move into estimations of (9). Alternatively, one can estimate (10) if there are stationary residuals underlying the several specifications of (8). Panel unit roots proposed by Levin, Lin and Chu (2002) and Im, Pesaran and Shin (2003) are reported in the bottom part with Schwarz criterion employed for lag-length selection. Alternative lag-length selection criteria, such as the Modified Akaike, did not change the qualitative results in any way. The first panel unit root test (henceforth LLC) assumes common unit root processes and the second (henceforth IPS) assumes individual unit root processes. The null hypothesis in all cases is that there is a unit root. Rejections of the null are highlighted in bold at the bottom of table 2. The tests convey a mixed result as the LLC usually rejects the null and suggest stationary residuals and the IPS yield nonstationary residuals except for model (8a) with WMXCH only as explanatory variable. If the residuals are calculated in first-differenced form, the null is rejected in all cases, which ensures stationary residuals. These results are not reported but are available upon request. Since there were mixed results across both tests, we proceed to analyze further the intermediate ADF statistics on the IPS testing procedure that assume each cross-section has a unique coefficient in the ADF testing. We thus report below the IPS test result in table 2 the average of the t-statistic across the 6 cross-section units and the maximum and minimum t-values. As can be seen, only (8a) with WMXCH provides relatively clustered values: maximum at -2.929 and minimum at -2.297. Everywhere else, there is much more variation across the t-values, suggesting at least that the common unit root assumption is

18

questionable. Therefore, based on the IPS testing procedure, we find support for stationary residuals only in specification (8a).8 Equation (9) was also estimated in first-differences and its results are omitted for space considerations. We estimate in sequence models with: 9a. only relative wages; 9b. wages and U.S. real output; and 9c. wages, U.S output and the NAFTA dummy. Real relative wages now do not affect employment negatively in first-differences. This result is observed no matter which of the two relative wages is employed. Specifications in firstdifferences in (9) have lower explanatory power overall, which is compatible with the (RHS) variables having no role in the explanation of the variance of changes in employment. The specifications are, however, clear of serial correlation problems according to the LM t-statistics and the Lagrange Multiplier NR2 statistics. As an alternative to (9), we put forward an error correction model (ECM) that combines the short-run properties of statistical relationships in first differenced form with the long-run properties of the model in level form. According to the Engle-Granger (1987) representation theorem, if the residuals on the benchmark model are stationary there is a linear combination of the non-stationary individual variables that is stationary. In that case, the estimates generated in level form in table 2 contain valuable information that can not be ignored. In order to explore the relationship between the long-run relationship that comes from the theory and the short-run adjustments through the ECM, equation (10) is estimated by three different methods in table 3. First, by FGLS methods as done before in models (8) and (9); second, by the FGLS with cross-section weights; and, third, by the SUR method which is appropriate when all the (RHS) regressors are exogenous and the errors are heteroskedastic and contemporaneous correlated. In each case the lagged residuals from each equation in levels (ECMit-1) are introduced into the (RHS) in an error-correction

19

fashion. It is then possible to see in table 3 that the β3 coefficient is found to be negative throughout: as deviations from the long-run equilibrium increase, the impact on the dependent variable decreases in an error-correction fashion. The values of the estimated coefficients on wages in table 3 suggest that MexicanChinese relative wages are an explanatory force of Mexican maquiladora employment dynamics in (10) with the value of -0.076 under FGLS-SUR. The point-value estimated is, however, small. A 10% increase in relative wages of Mexican with respect to Chinese manufacturing yields a 0.76% drop in employment. Contrary to speculation in the financial and popular press, there is no sound reason to believe that maquiladora employment levels fall substantially with higher (relative) Mexican wages. The results are only slightly higher in absolute value when modifications of the variance-covariance structure are assumed: in column (2) -0.090 for the FGLS-Cross Section estimation and in column (3) -0.091 if a SUR system based model is adopted. Estimated impacts of U.S. real output variations on employment are again strong and more than proportional, varying from 3.548 to 3.673 in the first set of columns. Other interesting findings are worth noting in the ECM results. First, the estimates show that between 30% and 33% of the lagged long-run equilibrium deviations are adjusted in the next period. This figure varies only slightly from the statistically significant β3’s in table 3 with FGLS (-0.302) to those estimated by SUR system methods (-0.335). Second, the strong and negative significance of the error correction term reinforces the idea that estimates generated in level form in table 2 contain valuable information that can not be ignored. In other words, the model linking employment to relative wages provides information that help explain the variation of maquiladora employment. Third, the explanatory power of the ECM-type models is fairly good, ranging from the 0.552 adjusted R2 of SUR to 0.996 of FGLS-SUR. A conservative figure should probably lie in between,

20

such as the adjusted R2 of 0.769 found by the FGLS under cross section weights. More importantly, the estimated coefficients of β1’s and β2’s are as expected: negative but small for Mexico-China relative wages and strongly positive for U.S. real output forces in the demand side.9 Fourth, note that the three columns at the right of table 3 report very similar results for the ECM if the residuals are taken from the FGLS with cross-section weights instead of SUR weighting. With the exchange rate and wage volatility of the 1990s and with Mexican maquiladoras located along the border being under intense competition, one would expect employment to react to such global forces. Results are weaker for the “Chinese competition hypothesis” than Gruben (2001)’s, leading support to the “entire package” approach discussed by Huro (2003), in which labor costs are just one factor among many possible that influence the foreign investor. The first-differenced model in Gruben (2001) explored the ratio of hourly Mexican manufacturing wages with respect to a group of 4 Asian countries. He obtained a statistically significant coefficient (at 10%) of about -0.11: as Mexican wages rise with respect to Asian, labor demand in the Mexican maquiladora falls slightly. The quantitative results from Gruben (2001) are thus very much in line with our β1 coefficients reported in table 3. According to our results, increases in U.S. real production consistently affect maquiladora employment positively. This happens across all specifications: in levels, in first-differences, and in ECM style. U.S. output fluctuations are thus the strongest factor behind maquiladora expansion. Increases in U.S. output have a more than three-fold impact on maquila employment growth: the response varies from 3.548 to 3.673 under WMXCH. These are larger than the coefficients in Gruben (2001) who presents a time series model for annual 1975-1999 with U.S. industrial production affecting positively

21

maquiladora’s aggregate employment. His U.S. industrial production coefficients ranged from 1.05 to 1.31. We conjecture that the stronger U.S. effects found herein are associated with the increasingly more integrated U.S. and Mexican economies in the 1990-2001 years than in the past. Uncovering more than proportional U.S. output effects on maquiladora employment dynamics, our two-step ECM procedure thus offers support to the anecdotal evidence based on the popular motto: “When the U.S. sneezes, Mexico catches a cold”.

22

5. Concluding Remarks Contrary to speculation in the press, this paper argues that there are no sound empirical arguments to justify that maquiladora employment levels fall substantially with higher (relative) Mexican wages. The only strong negative effects found in this paper are confined to estimations in levels. Under a two-step (panel) cointegration based procedure of first-differenced employment, the error-correction model leads to the effects of relative wages varying between -0.076 and -0.091, which are close to those reported by Gruben (2001). Overall, however, our results support stories of U.S. investors operating in a global world, such as Huro (2003) and Fullerton and Barraza de Anda (2003), in which labor costs are weighted against distance, infrastructure, and set up business costs. The very strong role of U.S. business cycles in the labor growth of the Mexican maquiladora industry is confirmed in this study. The effects of U.S. output growth are always more than proportional, with coefficients varying from 3.548 to 3.652. These estimated values are higher than those in Gruben (2001), whose focus was on the aggregate of Mexican maquiladoras (not a panel approach). The difference in results can be related to the closer integration between the two economies in the years 1990-2001. At least three topics seem noteworthy exploring. First, Braconier et al. (2005) show that MNEs react to relative wages (skilled to less skilled labor), following Berman et al. (1998) and Machin and Van Reenen (1998) and studies on skill biased technical change (SBTC). Second, extensions of the theoretical model to capital and labor market imperfections, as emphasized by Cypher (2001) and Kopinak (1996), could lead to further hypotheses testing. Third, dynamic panel methods could complement this paper, particularly with respect to the simultaneity between employment and wages. Critical in this case are a general equilibrium model and the choice of good instrumental variables, as in Feenstra and Hanson (1997). These extensions are left for further research.

23

References Banamex, 2003. The Maquiladora: A Changing Industry. Review of the Economic Situation of Mexico. Economic and Social Research, No. 936, Vol. 79, 499-503. Berman, Eli, Bound, John, Machin, Stephen, 1998. Implications of Skill-Biased Technological Change: International Evidence. Quarterly Journal of Economics 113 (4), 1245-1279. Bernard, Andrew, Jensen, J. Bradford, Schott, Peter, 2005. Survival of the Best Fit: Exposure to Low Wage Countries and the (Uneven) Growth of U.S. Manufacturing Plants. Journal of International Economics, forthcoming. Braconier, Henrik, Norback, Pehr-Johan, Urban, Dieter, 2005. Multinational Enterprises And Wage Costs: Vertical FDI Revisited. Journal of International Economics, forthcoming. Cypher, James, 2001. Developing Disarticulation within the Mexican Economy. Latin American Perspectives 28 (3), 11-37. Dunning, John, 1988. Explaining International Production. University Hyman, London. The Economist, 2003. China’s Economy: Is the Wakening Giant a Monster? The Economist, Feb 13, 2003. Engle, Robert, Clive Granger, 1987. Cointegration and Error Correction: Representation, Estimation, and Testing. Econometrica 55 (2), 251-276. Feenstra, Robert, Hanson Gordon, 1997. Foreign Direct Investment and Relative Wages: Evidence from Mexico’s Maquiladoras. Journal of International Economics 42, 371-393. Fullerton, Thomas, Barraza de Anda, Martha, 2003. Maquiladora Prospects in a Global Environment. Texas Business Review, October 1-5. Fullerton, Thomas, Schauer David, 2001. Short-Run Maquiladora Employment Dynamics. International Advances in Economic Research 7, 471-478. Gruben, William, 2001. Was NAFTA Behind Mexico’s High Maquiladora Growth?, Economic and Financial Review, FRB of Dallas, Third Quarter, 11-21. Hanson, Gordon, 2002. The Role of Maquiladoras in Mexico’s Export Boom. University of California-San Diego, manuscript. Hanson, Gordon, 2001. U.S. – Mexico Integration and Regional Economies: Evidence from Border-City Pairs. Journal of Urban Economics 50, 259-287. Hsiao, Cheng, Pesaran, M. Hashem, Tahmiscioglu, A. Kamil, 2002. Maximum Likelihood Estimation of Fixed Effects Dynamic Panel Data Models Covering Short Time Periods. Journal of Econometrics 109, 107-150.

24

Hummels, David, Ishii, Jun, Yi, Kei-Mu, 2001. The Nature and Growth of Vertical Specialization in World Trade. Journal of International Economics 54, 75-96. Huro Michelini, Doreen, 2003, Mexico vs. China: The Debate Continues. Mexico Now May-June 2003, 38-41. Im, Kyung So, Pesaram, M. Hashem, Shin, Yongcheol, 2003. Testing for Unit Roots in Heterogenous Panels. Journal of Econometrics 115 (1), 53-74. Kopinak, Kathryn, 1996. Desert Capitalism: Maquiladoras in North America’s Western Industrial Corridor. University of Arizona Press, Tucson. Kose, M. Ayhan, Meredith, Guy, Towe, Christopher, 2004. How has NAFTA Affected the Mexican Economy? Review and Evidence. (April). IMF Working Paper. Levin, Andrew, Lin, Chien-Fu, Chu, Chia-Shang James, 2002. Unit Roots in Panel Data: Asymptotic and Finite Sample Properties. Journal of Econometrics 108, 1-24. Machin, Stephen, Van Reenen, John, 1998. Technology and Changes in Skill Structure: Evidence from Seven OECD Countries. Quarterly Journal of Economics 113 (4), 1215-1244. Markusen, James, 1995. The Boundaries of Multinational Entreprises and the Theory of International Trade. Journal of Economic Perspectives 9 (2), 169-189. Milner, Chris, Wright, Peter, 1998. Modeling Labour Market Adjustment to Trade Liberalisation in an Industrializing Economy. Economic Journal 108, 509-528. Milwaukee Journal Sentinel, 2003. Mexico Losing NAFTA Advantage: Country’s Economy Languishing as Other Nations Lure Away Investment. JSOnline, November 27, 2003. Mollick, André, 2003. Employment Determination at Mexican Maquiladoras: Does Location Matter? Journal of Borderlands Studies 18 (2), 45-67. Mollick, André, Cortez, Abigail, Olivas, Rosa, 2005. Local Labor Markets in U.S. -Mexican Border Cities and the Impact of Mexican Maquiladora Production. Annals of Regional Science, forthcoming. Mortimore, Michael, Calderón, Alvaro, Moussa, Nicole, 2003. Foreign Investment in Latin America and the Caribbean. United Nations, ECLAC. Nation & World, 2003. Mexico Fears Competition from China: Imported Goods Winning Trade War. The Morning News 3B, November 24, 2003. Robertson, Raymond, 2003. Exchange Rates and Relative Wages: Evidence from Mexico. The North American Journal of Economics and Finance 14, 25-48. Slaughter, Matthew, 2001. International Trade and Labor-Demand Elasticities. Journal of

25

International Economics 54, 27-56.

26

Figure 1. Mexico-China Relative Wages (WMXCH) and the MXN Multilateral Real Exchange Rate Relative Mexican/Chinese Wages (right axis) and the MXN Multilateral Real Exchange Rate (q) 140

30

120

25

100 20 80 15 60 10 40

5

20

0

0 1990

1991

1992

1993

1994

1995 q

1996 wmxch

1997

1998

1999

2000

2001

27

Figure 2. Mexico-U.S. Relative Wages (WMXUS) and the MXN Multilateral Real Exchange Rate Relative Mexican/U.S. Wages (right axis) and the MXN Multilateral Real Exchange Rate (q) 140

16

14

120

12 100 10 80 8 60 6 40 4

20

2

0

0 1990

1991

1992

1993

1994

1995 q

1996 wmxus

1997

1998

1999

2000

2001

28

Figure 3. U.S. Variables: Real GDP (Y*, right axis) and 12-month Nominal T-bill Interest Rates (left axis) U.S. Real GDP (y*) in USD billion and U.S. 12-month T-bill rate (i*) in % 9

10000

8

9000

8000

7

7000 6 6000 5 5000 4 4000 3 3000 2

2000

1

1000

0

0 1990

1991

1992

1993

1994

1995

1996 i*

y*

1997

1998

1999

2000

2001

29

Figure 4. Maquiladora’s Employment across Mexican Border States. Maquiladora Employment (L: Number of Employees) at Mexican Border States 350,000

300,000

250,000

200,000

150,000

100,000

50,000

1990

1991

1992 Border L_bc

1993

1994 Border L_chi

1995 Border L_coa

1996 Border L_nl

1997

1998

Border L_son

1999 Border L_tam

2000

2001

30

Table 1. Comparison of Earnings in Manufacturing between Mexico, China and the U.S. 1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

China Manufacturing wages*

172.25

190.75

219.58

279.00

356.92

430.75

470.17

494.42

588.67

649.50

729.17

814.50

CPI base 1990

100.00

105.10

113.20

132.50

165.60

193.50

209.50

215.40

213.70

210.70

210.90

212.40

(W/P)CH

172.25

181.49

193.98

210.57

215.53

222.61

224.42

229.54

275.47

308.26

345.74

383.47

4.78

5.32

5.51

5.76

8.62

8.35

8.31

8.29

8.28

8.28

8.28

8.28

36.01

34.09

35.17

36.54

25.01

26.66

26.99

27.69

33.27

37.24

41.76

46.33

Yuan/USD (W/P)CH in USD

Mexico Manufacturing wages** 1666.64 2190.16 2714.63 3108.96 3453.12 3610.79 4404.45 5276.09 6288.08 7437.91 8621.89 9763.60 CPI base 1990 (W/P)MX MXN/USD (W/P)MX in USD (W/P)MX in USD / (W/P)CH in USD or WMXCH WMXUS

100.00

122.70

141.70

155.50

166.30

224.50

301.70

364.00

422.00

491.90

538.60

572.90

1666.64 1784.97 1915.76 1999.33 2076.44 1608.37 1459.88 1449.48 1490.07 1512.08 1600.80 1704.24 2.81

3.02

3.09

3.12

3.38

6.42

7.60

7.92

9.14

9.56

9.46

9.34

592.56

591.36

619.01

641.71

615.22

250.55

192.10

183.05

163.10

158.16

169.30

182.42

16.45

17.35

17.60

17.56

24.60

9.40

7.12

6.61

4.90

4.25

4.05

3.94

11.00

12.33

13.67

15.00

15.00

10.00

9.00

9.00

9.00

10.00

11.00

11.00

Source: When calculating WMXCH, Chinese wage data are collected from ILO and Mexican wage data are from INEGI´s Monthly Industrial Survey. WMXUS is taken directly from U.S. BLS publication “Hourly Compensation Costs for Production Workers in Manufacturing.” The WMXUS figures for 1991 and 1992 were linearly interpolated; all other figures come from the BLS publication. In the calculations for WMXCH, local currencies are: yuan for China and nuevos pesos for Mexico; the spot annual average of monthly exchange rates are used for conversion into U.S. dollars. * Average earnings of the month for all type of manufacturing employees in local currency. ** Average monthly salary by worker of manufacturing.

31

Table 2. Panel Data FGLS Estimations of Maquiladora Employment in Levels. logLit = α + β0i + β1logWt + εit

(8a)

logLit = α + β0i + β1logWt + β2logY*t + εit

(8b)

logLit = α + β0i + β1logWt + β2logY*t + β3NAFTA + εit

(8c)

With WMXCH β1

-0.496*** (0.050)

β2

With WMXUS -0.039 (0.025)

-0.027 (0.023)

2.979*** (0.154)

2.875*** (0.204)

β3

1.567

LM t-stat

7.076***

Adj. R2

-0.124*** (0.043)

-0.072** (0.034)

3.108*** (0.071)

3.016*** (0.091)

0.139*** (0.026)

DW

LM NR2 stat

-0.358** (0.176)

0.110*** (0.019)

1.506

1.667

1.600

1.595

1.670

7.203***

7.424***

13.537***

6.997***

8.089***

30.756**

32.802**

41.316**

52.80**

29.106**

43.362**

0.9996

0.9999

0.99998

0.9995

0.99996

0.99998

-4.770 [0.000] -2.586 [0.005]

-3.057 [0.001] -0.528 [0.299]

-2.133 [0.017] -0.738 [0.230]

-1.090 [0.138] 1.949 [0.974]

-3.447 [0.000] -0.911 [0.181]

-2.186 [0.014] -0.706 [0.240]

Panel Unit Root Tests on Residuals LLC IPS

Average IPS -2.583 -1.727 -1.818 -0.678 -1.889 -1.804 Values of tstatistics Maximum t-2.297 0.646 -0.337 -0.248 0.242 -0.302 stat. Values Minimum t-2.929 -3.479 -2.635 -1.184 -3.549 -2.681 stat. Values Notes: i. Data are of annual frequency from 1990 to 2001 per state. The six states of the border comprehend Baja California, Chihuahua, Coahuila, Nuevo León, Sonora and Tamaulipas. The number of observations is 72. ii. Below the coefficients are White-diagonal standard errors (corrected for degrees of freedom). iii. The method of estimation is the fixed effects model, which includes a constant term (α) and fixed effects terms (β0i) that differ across states. We estimate by SUR weights, which is the feasible estimator when the residuals are both cross-section heteroskedasticity and contemporaneously correlated. iv. The LM t-stat. is the t-statistic associated with the lagged residual within a standard Lagrange Multiplier test on the residuals of the regression. The LM NR2 stat. is the value derived from N and R2 computed in this auxiliary regression. The statistic has a chi-squared distribution with degrees of freedom equal to the number of estimated parameters (q): χ2 (2) = 5.99; χ2 (3) = 7.81; and χ2 (4) = 9.49. The LM NR2 stat. is calculated under the null of no serial correlation up to lag order 1, which is reasonable for annual data. v. The symbols *, **, and *** refer to levels of significance of 10%, 5%, and 1%, respectively. vi. Panel unit roots proposed by Levin, Lin and Chu (LLC) (2002) and Im, Pesaran and Shin (IPS) (2003) are reported in the bottom with the Schwarz criterion used for lag-length selection (p-values in parenthesis). LLC assumes common unit root processes and IPS assumes individual unit root processes. The null hypothesis is that there is a unit root; rejections of the null are highlighted in bold.

32

Table 3. Panel Data FGLS and SUR Estimations of Maquiladora Employment: ECM-Type based on Residuals of (8a). ∆logLit = α + β0i + β1∆logWXMCHt + β2∆logY*t + β3ECMit-1 + εit

(10)

Residuals from (8a) by FGLSCross Section Weights

Residuals from (8a) by FGLSSUR Weights

SUR System

FGLSSUR Weights

FGLSCross Section Weights

SUR System

-0.090*** (0.016)

-0.091*** (0.008)

-0.087*** (0.005)

-0.095*** (0.014)

-0.096*** (0.008)

3.548*** (0.050)

3.673*** (0.302)

3.652*** (0.135)

3.523*** (0.062)

3.622*** (0.289)

3.596*** (0.136)

-0.302*** (0.006)

-0.320*** (0.039)

-0.335*** (0.016)

-0.332*** (0.007)

-0.331*** (0.038)

-0.348*** (0.016)

2.355

2.490

2.260

2.195

2.507

2.278

-2.107**

-1.990*

-2.045**

-2.113**

LM NR2 stat

5.52

4.08

4.56

4.38

Adj. R2

0.996

0.769

0.993

0.783

FGLSSUR Weights

FGLSCross Section Weights

β1

-0.076*** (0.003)

β2

β3

DW LM t-stat

0.552

0.558

Notes: i. Data are of annual frequency from 1990 to 2001 per state. The six states of the border comprehend Baja California, Chihuahua, Coahuila, Nuevo León, Sonora and Tamaulipas. The number of observations is 66. ii. ECM is the lagged one period residuals from the equations in levels reported in table 2 (first three columns) and for the equations in levels calculated (last three columns). iii. Below the coefficients are standard errors. iv. The methods of estimations are as follows: the feasible generalized least squares (FGLS) with SURweights, FGLS with cross-section weights, and system-based SUR methods. In all cases, constant term (α) and fixed effects terms (β0i) that differ across states are included. v. For the FGLS estimations, we construct the LM t-statistic associated with the lagged residual within a standard Lagrange Multiplier test on the residuals of the regression. Also, the LM NR2 stat. is the value derived from N and R2 computed in this auxiliary regression. The statistic has a chi-squared distribution with degrees of freedom equal to the number of estimated parameters (q): χ2 (4) = 9.49. The LM NR2 stat. is calculated under the null hypothesis of no serial correlation up to lag order 1, which is reasonable for annual data. For the SUR system estimation, we average the six values of the DW and the Adj. R2 statistics. vi. The symbols *, **, and *** refer to levels of significance of 10%, 5%, and 1%, respectively.

33

ENDNOTES 1

Very illustrative is the testimony of a U.S.-based executive and V.P. of Global

Operations for Dial Tool Industries Inc., Ms. Doreen Huro, who travels often to Mexico and China. In Huro (2003), a wide-ranging list of pros and cons of moving operations from Mexico to China is provided. She mentions that, at USD 0.50 per hour, no doubt China has lower labor costs than Mexico. But what matters for decision-making executives is the entire package. Weighting against the Chinese option are, for example, utilities (electricity and phone lines) in China that are well below Mexican costs, but in many areas of the country they are not reliable and are of poor quality. Travel time from Chicago varies from 4-5 hours to Mexico to 19 or more hours to China. The time difference plays some part as well if constant communications and support are needed in the day-to-day business. Consolidated ocean fright delivers a clear trade-off in the 4-5 days travel time from Mexico and 4-5 weeks in China. Finally, compared to China, start up in Mexico is a relatively painless process. She mentions, on the other hand, that Mexico needs to improve on the charges imposed by Import/Export Brokers, which are usually higher (USD 100) against those by an average broker in China. Also, although overtime in China is paid similar to Mexico, the willingness to work extra hours in China surpasses the enthusiasm, or lack of, in Mexico. 2

Manufacturing wage comparisons across countries are clearly the most accurate.

Besides, maquiladoras are typically associated with manufacturing activities, at least in the traditional sense. As far as the maquiladora real effects on the economy are concerned, however, evidence varies. While empirical work has shown that maquiladora production have exerted substantial effect on manufacturing in U.S. cities, as documented by Hanson (2001), more recent evidence shows that maquiladora value added growth has had more impact from 1990 onwards on sectors other than manufacturing in U.S. border

34

cities. See Mollick et al. (2005) for a time series approach to the Brownsville-Matamoros and El Paso-Juárez city pairs. 3

In Mollick (2003), Mexican wages are separated into: wages paid to non-production

workers (“Empleados”, Ws) and wages paid to production workers (“Obreros”, Wu). Wages are then deflated by the consumer price index (CPI) of the state’s main city, also from INEGI. Here we do not have Chinese data to do the same when calculating WMXCH. Rather, the manufacturing Chinese wage data from ILO includes both administrative and assembly-line workers, weighted accordingly. Wages across Mexican states in Mollick (2003) indicate stable real wages paid by the maquiladora industry with a fall in 1995 when the Mexican economy collapsed after the December of 1994 currency crash. As shown in figures 1 and 2 and compared to Mexican real wages at the state level, WMXCH and WMXUS present much more fluctuations over time, as these values include exchange rate effects. 4

Several studies of the Mexican economy suggest Mexico operates under less than

optimal capital and labor market conditions. Cypher (2001, pp 27-28) develops a disarticulation thesis on the Mexican economy, in which conglomerates, sometimes with significant foreign ownership, are able to obtain credit from the Mexican financial system, the Mexican government, and international lenders. Kopinak (1996) emphasizes that the maquiladora labor market is segmented along both gender and skill lines. She also mentions important information asymmetry problems between maquiladora workers and managers. Incorporating these factors to the analysis is beyond the scope of the paper. What can be said, however, is that the recent U.S. recession can be associated with a decrease in the number of establishments. As of January of 2005, there are 2,811 maquiladora firms operating in Mexico, of which 2,065 are located in states along the U.S.-Mexico border (ratio of 73.46%). At the time of the start of our econometric

35

analysis (January of 1990), there were 1,594 maquiladora firms, of which 1,376 (ratio of 86.32%) were located at the border. At the end of the analysis (December of 2001), there were 3,279 maquiladora firms, of which 2,811 (ratio of 72.43%) were located at the border. This information is from INEGI’s BIE. These data support more concentration within the maquiladora industry when the U.S. economy is in the downturn phase. 5

Previous versions of this paper used either maquiladora output or state GDP as the

demand variable in the econometric model. External variables, such as the real exchange rate (q) in Robertson (2003, p. 27), are known to exert effects on the stable relationship between employment and firm’s output and costs. Here, however, exchange rate effects are already included in WMXCH and in WMXUS. 6

In panel data, researchers allow for “individual” (in our case, Mexican state-specific)

and time-invariant effects. As noted by Hsiao et al. (2002, pp. 107-108), “this approach leads to treating such effects as additional parameters to be estimated along with the other parameters that are of more substantive interest. In the case when the time dimension, T, is fixed, the introduction of individual effects increases the number of parameters to be estimated with the increase in the number of observations on the cross-sectional dimension, N. The dynamic nature of the model also gives rise to the ‘initial value’ problem.” Because the estimation of individual effects is not asymptotically independent of the estimation of structural parameters, the maximum likelihood estimator (MLE) is inconsistent if the time-series dimension is finite. Hsiao et al. (2002) show that there exists a linear transformation that eliminates the individual effects and that maximization of the transformed likelihood function can yield consistent estimators when N→∞, independent of whether T is fixed or T→∞. 7

We conduct two sorts of serial correlation tests, both derived from the Lagrange

Multiplier (LM) Breusch-Godfrey test. For each panel equation we regress the computed

36

residuals on the right hand side variables and on (lagged one period) residuals. In the annual data context, any serial correlation is likely to appear with only one lagged residual term. The tables below report the LM t-statistics on the lagged residual term and the LM NR2 statistic, which has a χ2(p) distribution, where p is the number of parameters in the auxiliary regression. 8

Other panel unit root tests do support this finding. The panel unit root results remain

unchanged when FGLS with cross-section weights are employed instead of the SURweights of table 2. 9

Note that relative real wages impact employment negatively under (10), whose

benchmark long-run model provide stationary residuals according to both LLC and IPS tests. In that case, the ECM has all the conditions required since all regressors are stationary. We also checked the stationary residuals of each equation in table 2: 8a, 8b and 8c under both WMXCH and WMXUS into (10). Results vary, particularly with respect to the wage variable as the β1 coefficient on wages is found to be either small negative or not statistically significant at all. In any case, this set of estimates hinges on the stationary assumptions for the residuals of the benchmark model, which clearly was not the case according to the LLC and IPS tests. This set of estimates is available from the authors.